CN110334728B - Fault early warning method and device for industrial internet - Google Patents
Fault early warning method and device for industrial internet Download PDFInfo
- Publication number
- CN110334728B CN110334728B CN201910372649.9A CN201910372649A CN110334728B CN 110334728 B CN110334728 B CN 110334728B CN 201910372649 A CN201910372649 A CN 201910372649A CN 110334728 B CN110334728 B CN 110334728B
- Authority
- CN
- China
- Prior art keywords
- early warning
- class
- prediction
- model
- industrial equipment
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B15/00—Systems controlled by a computer
- G05B15/02—Systems controlled by a computer electric
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/06—Management of faults, events, alarms or notifications
Abstract
The invention discloses a fault early warning method and device for industrial internet. The method comprises the following steps: correcting the early warning threshold value of the early warning model according to early warning information sent by the early warning model, actual faults of the industrial equipment and actual production stoppage; the adjustment formula of the early warning threshold XT is as follows: XT ═ (X2+ X3)/4+ (X4 × W4+ X1 × W1)/2 where XT represents an early warning threshold; x1, X2, X3 and X4 respectively represent a centroid value of a prediction error class, a centroid value of an accurate prediction fault class, a centroid value of an accurate prediction normal class and a centroid value of a prediction inaccuracy class; w1 and W4 denote a weight coefficient of the prediction error class and a weight coefficient of the prediction inaccuracy class, respectively. The method effectively balances the fault early warning and the production efficiency, and ensures the safe and efficient operation of industrial equipment.
Description
Technical Field
The invention relates to the technical field of industrial internet, in particular to a fault early warning method and device for industrial internet.
Background
In the industrial production process, the operation condition of industrial equipment needs to be monitored so as to find and eliminate problems in time. With the further industrialization, people put forward higher requirements on the operation efficiency of industrial equipment, hope to realize monitoring and early warning on the industrial equipment and reduce the production stagnation caused by the failure of the industrial equipment.
The early warning mode adopted at present is an edge model deployment mode, namely a locally trained fault early warning model is deployed on an edge cloud, then production data is accessed in real time, and fault early warning is realized according to the real-time production data. The early warning method has the following defects:
firstly, the edge model is completely dependent on the model at the initial operation stage, but the edge model has less actual data at the initial operation stage, and prediction errors are easily caused by completely depending on the model.
Secondly, the threshold value of the edge model early warning is set according to historical experience, if the predicted failure probability is set to be greater than 0.6, the industrial equipment is considered to be failed, and the prediction is prone to be inaccurate due to the cutting mode.
And thirdly, when the early warning model is trained, the balance between early warning and productivity is not considered, the training and updating are carried out only by the error of the edge model, the core purpose of neglecting industrial equipment is high efficiency and high yield, and the early warning is disconnected with the productivity.
Therefore, the current early warning model has prediction errors, namely faults occur but are not predicted, and unnecessary loss is caused; or, the early warning model has inaccurate prediction, namely, the prediction is failed, but no fault exists actually, so that the industrial equipment is stopped to carry out useless maintenance, the use efficiency of the industrial equipment is not high, and the loss is caused to enterprises. In other words, the current early warning model cannot effectively balance the early warning with the productivity, which results in a wrong prediction or an inaccurate prediction of the industrial equipment.
Disclosure of Invention
Therefore, the invention provides a fault early warning method and device for an industrial internet, and aims to solve the problem that the use efficiency of industrial equipment is low due to unreasonable early warning models in the prior art.
In order to achieve the above object, a first aspect of the present invention provides a fault early warning method for an industrial internet, including:
correcting the early warning threshold value of the early warning model according to early warning information sent by the early warning model, actual faults of the industrial equipment and actual production stoppage;
the adjustment formula of the early warning threshold XT is as follows:
XT＝(X2+X3)/4+(X4×W4+X1×W1)/2
wherein XT represents an early warning threshold;
x1, X2, X3 and X4 respectively represent a centroid value of a prediction error class, a centroid value of an accurate prediction fault class, a centroid value of an accurate prediction normal class and a centroid value of a prediction inaccuracy class;
w1 and W4 respectively represent the weight coefficient of the prediction error class and the weight coefficient of the prediction inaccuracy class; W4/W1 ═ a, and W1+ W2 ═ 1;
a is a weight adjustment constant, and a is SumC1/SumC 4;
wherein SumC1 represents the sum of the downtime of the misprediction class;
SumC4 represents the sum of the downtime of the predicted inaccurate class.
Wherein, include:
training a basic model according to historical production information of industrial equipment to obtain a training model;
testing the training model to obtain an early warning model, and determining an early warning threshold value in the early warning model;
early warning actual operation data of the industrial equipment by using the early warning model with the early warning threshold value;
recording the early warning sent by the early warning model, the actual fault and the actual production stop time of the industrial equipment;
and correcting the early warning threshold value of the early warning model according to the early warning information sent by the early warning model, the actual fault of the industrial equipment and the actual production stopping time.
Wherein the step of training the model to obtain the training model according to the historical production information of the industrial equipment comprises:
acquiring historical production information of industrial equipment;
extracting input data and output data from the historical production information;
configuring the input data and the output data to an original model;
and training the original model according to the input data and the output data to obtain the training model.
Wherein the input data comprises: production capacity per unit time, temperature of industrial equipment operation and power consumption;
the output data includes: the data is used for representing the data before the industrial equipment is in failure and representing the data that the industrial equipment is in normal operation.
Wherein, the step of correcting the early warning threshold value of the early warning model according to the early warning information sent by the early warning model, the actual fault and the actual production halt of the industrial equipment comprises the following steps:
calculating the sum of the production stopping time of the prediction error class SumC1 and the sum of the production stopping time of the prediction inaccuracy class SumC4 according to the actual production stopping time;
calculating a weight adjustment constant a according to the sum of the production stop time SumC1 of the prediction error class and the sum of the production stop time SumC4 of the prediction inaccuracy class;
obtaining a weight coefficient W1 of a prediction error class and a weight coefficient W4 of a prediction inaccuracy class according to the weight adjusting constant a;
performing cluster analysis according to the early warning information sent by the early warning model and the actual fault of the industrial equipment to obtain a mass center value X1 of a prediction error class, a mass center value X2 of an accurate prediction fault class, a mass center value X3 of an accurate prediction normal class and a mass center value X4 of a prediction inaccurate class;
and correcting the early warning threshold value of the early warning model according to the cluster analysis result, the weight coefficient W1 of the prediction error class and the weight coefficient W4 of the prediction inaccuracy class.
Wherein, in the step of clustering calculation according to the early warning information sent by the early warning model and the actual fault of the industrial equipment,
and if the mass center value X4 of the inaccurate prediction class is smaller than the mass center value X1 of the wrong prediction class, or the mass center value X3 of the normal prediction class is smaller than the mass center value X1 of the wrong prediction class, the early warning model is considered to be wrong, and the early warning model is reconstructed.
A second aspect of the present invention provides an industrial internet-oriented fault early warning apparatus, including:
the training model establishing unit is used for training a basic model according to historical production information of the industrial equipment to obtain a training model;
the early warning threshold establishing unit is used for testing the training model to obtain an early warning model and determining an early warning threshold in the early warning model;
the early warning unit is used for early warning actual operation data of the industrial equipment by using the early warning model with the early warning threshold value;
the recording unit is used for recording the early warning sent by the early warning model, the actual fault and the actual production stop time of the industrial equipment;
the early warning threshold correction unit is used for correcting the early warning threshold of the early warning model according to early warning information sent by the early warning model, actual faults and actual production stoppage of the industrial equipment;
the adjustment formula of the early warning threshold value is as follows:
XT＝(X2+X3)/4+(X4×W4+X1×W1)/2
wherein XT represents an early warning threshold;
x1, X2, X3 and X4 respectively represent a centroid value of a prediction error class, a centroid value of an accurate prediction fault class, a centroid value of an accurate prediction normal class and a centroid value of a prediction inaccuracy class;
w1 and W4 respectively represent the weight coefficient of the prediction error class and the weight coefficient of the prediction inaccuracy class; W4/W1 ═ a, and W1+ W2 ═ 1;
a is a weight adjustment constant, and a is SumC1/SumC 4;
wherein SumC1 represents the sum of the downtime of the misprediction class;
SumC4 represents the sum of the downtime of the predicted inaccurate class.
Wherein the training model establishing unit includes:
the production information acquisition module is used for extracting historical production information of the industrial equipment;
the extraction module is used for extracting input data and output data from the historical production information;
a configuration module for configuring the input data and output data to the base model;
and the training module is used for training the basic model according to the input data and the output data to obtain the training model.
Wherein, the early warning threshold value correcting unit comprises:
the production stop time calculation module is used for calculating the sum of the production stop times SumC1 of the prediction error class and the sum of the production stop times SumC4 of the prediction inaccuracy class according to the actual production stop time;
the weight adjusting constant calculating module is used for calculating a weight adjusting constant a according to the sum of the production stopping time of the prediction error class SumC1 and the sum of the production stopping time of the prediction inaccuracy class SumC 4;
the weight coefficient calculation module is used for obtaining a weight coefficient W1 of the prediction error class and a weight coefficient W4 of the prediction inaccuracy class according to the weight adjusting constant a;
the cluster calculation module is used for carrying out cluster calculation according to the early warning information sent by the early warning model and the actual faults of the industrial equipment to obtain a mass center value X1 of a predicted error class, a mass center value X2 of an accurate predicted fault class, a mass center value X3 of an accurate predicted normal class and a mass center value X4 of an inaccurate predicted class;
and the threshold correction module is used for correcting the early warning threshold of the early warning model according to the cluster analysis result, the weight coefficient W1 of the prediction error class and the weight coefficient W4 of the prediction inaccuracy class.
Wherein, in the cluster calculation module,
and if the mass center value X4 of the inaccurate prediction class is smaller than the mass center value X1 of the wrong prediction class, or the mass center value X3 of the normal prediction class is smaller than the mass center value X1 of the wrong prediction class, the early warning model is considered to be wrong, and the early warning model is reconstructed.
The invention has the following advantages:
according to the fault early warning method for the industrial internet, when the early warning threshold value of the early warning model is adjusted, cluster analysis is carried out on the prediction result of the early warning model and the actual operation condition of the industrial equipment to obtain the centroid value X1 of a prediction error class, the centroid value X2 of an accurate prediction fault class, the centroid value X3 of an accurate prediction normal class and the centroid value X4 of an inaccurate prediction class, meanwhile, the weights of the production stop time caused by the actual fault of the industrial equipment and the production stop time caused by the inaccurate prediction of the early warning model are comprehensively considered, the accuracy of early warning is improved, the production stop time caused by the inaccurate prediction is reduced, the production efficiency is improved, meanwhile, unnecessary loss caused by the error of prediction is reduced, the fault early warning and the production efficiency are effectively balanced, and the industrial equipment can safely and efficiently operate; meanwhile, a foundation is laid for realizing real-time updating and standardization flow of decisions.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention.
Fig. 1 is a flowchart of a fault early warning method for an industrial internet according to an embodiment of the present invention;
FIG. 2 is a flow chart of obtaining a training model according to an embodiment of the present invention;
FIG. 3 is a cluster analysis diagram after a test result obtained by testing a training model is subjected to cluster analysis by using a test set according to an embodiment of the present invention, wherein an abscissa X represents a probability of predicting a fault, and an ordinate Y represents a probability of actually occurring the fault;
FIG. 4 is a flowchart of adjusting an early warning threshold according to an embodiment of the present invention;
fig. 5 is a schematic block diagram of a fault early warning device for an industrial internet according to an embodiment of the present invention;
FIG. 6 is a schematic block diagram of a training model building unit in an embodiment of the present invention;
FIG. 7 is a schematic block diagram of an early warning threshold correction unit according to an embodiment of the present invention;
fig. 8 is a flowchart illustrating the overall operation of the fault warning device in the embodiment of the present invention.
In the drawings:
1: training model establishing unit 2: early warning threshold value establishing unit
3: the early warning unit 4: recording unit
5: early warning threshold value correcting unit 11: production information acquisition module
12: the extraction module 13: configuration module
14: the training module 51: production stop time calculation module
52: weight adjustment constant calculation module 53: weight coefficient calculation module
54: cluster calculation module 55: early warning threshold correction module
81: the fault early warning device 82: industrial equipment
83: underlying databases
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating the present invention, are given by way of illustration and explanation only, not limitation.
The embodiment provides a fault early warning method for industrial internet. As shown in fig. 1 and 2, the method includes:
and step S1, training the basic model according to the historical production information of the industrial equipment to obtain a training model.
As shown in fig. 2, step S1 specifically includes:
step S11, historical production information of the industrial equipment is acquired.
The historical production information of the industrial equipment includes information that the industrial equipment has a fault during operation and information that the industrial equipment has not a fault (a fault is manually predicted but actually does not have a fault).
In step S12, input data and output data are extracted from the historical production information.
The input data includes, but is not limited to, production capacity per unit time, temperature of operation of the industrial equipment, power consumption, etc. In order to improve the training efficiency of the training model, the time window of the input data is the data of five minutes before the industrial equipment is in fault and is not in fault.
The output data includes: the data is used for representing the data before the industrial equipment is in failure and representing the data that the industrial equipment is in normal operation. In this embodiment, "1" represents data before the occurrence of the failure of the industrial equipment, and "0" represents data in which the industrial equipment operates normally.
Step S13, the input data and the output data are configured to the original model.
The original model is an entropy Markov model, and input data and output data are configured to the original model.
And step S14, training the original model according to the input data and the output data to obtain a training model.
And configuring input data and output data in the original model to carry out model training, wherein the data result is a trained training model. Real-time production data is used as input data and input into the training model, and the probability of fault occurrence after N minutes can be predicted.
It should be noted that, during model training, input data and output data are divided into a training set and a test set according to a certain proportion, and the training set establishes a mapping relationship between production data and 0 or 1 through an algorithm. The test set is a set for judging whether the mapping relation is accurate or not, and the accuracy of the training model is estimated through indexes of the test set.
And step S2, testing the training model to obtain an early warning model, and determining an early warning threshold value in the early warning model.
And testing the training model by using the test set, predicting the probability of possible failure of the industrial equipment, wherein the probability is from 0 to 1, and the larger the numerical value is, the higher the possibility of the failure of the industrial equipment is.
And performing cluster analysis on the test result and the actual data of the test set by adopting a K-Means clustering method. As shown in fig. 3, the clustering results are classified into four classes, where the first class C1 represents a prediction error class, that is, the training model predicts that the industrial equipment will not be out of order, but actually has a failure; the second type C2 represents an accurate prediction fault type, namely, the training model predicts that the industrial equipment will be in fault and actually has fault; the third class C3 represents the accurate prediction normal class, i.e., the training model predicts that the industrial equipment will not fail, and there is actually no failure; the fourth class C4 represents the prediction inaccuracy class, i.e., the training model predicts that the industrial equipment will fail, but in fact there is no failure (false alarm).
The centroid value (center value) of each class, that is, the median of the X-axis coordinate values corresponding to all prediction points in each class (dot-dash line in fig. 3) is calculated, and the center points of class C1 to class C4 are denoted as (X1,1), (X2,1), (X3,0), and (X4,0), respectively.
And determining an early warning threshold according to the centroid value, wherein when the probability is greater than the early warning threshold, more points are contained in the categories C2 and C3, and less points are contained in the categories C1 and C4.
The early warning threshold value is (X4+ X1)/2 as the average value of the centroid values of the category C1 and the category C4, and (X2+ X3)/2 as the average value of the centroid values of the category C2 and the category C3, and the central value of the two average values is the early warning threshold value XT:
XT＝(X2+X3)/4+(X4+X1)/4
it should be noted that, when X4 < X1, or X3 < X1, it indicates that the training model trained in step S1 is erroneous, and step S1 needs to be executed again.
And step S3, early warning actual operation data of the industrial equipment by using an early warning model with an early warning threshold value.
And deploying the prediction model to a production system, and docking real-time production data.
And step S4, recording the early warning sent by the early warning model, the actual fault of the industrial equipment and the actual production stop time.
When the early warning model monitors the operation data of the actual industrial equipment to send out the early warning, the early warning is recorded, the operation data five minutes before the early warning is sent out is sent to maintenance management personnel, and the maintenance management personnel judge whether the production is stopped for maintenance or not according to the data of the five minutes and the experience. If the judgment is consistent with the early warning, stopping production and overhauling; if the judgment is inconsistent with the early warning, the maintenance is not stopped to improve the production efficiency of the industrial equipment.
After the maintenance personnel maintain the machine, the maintenance results, such as successful maintenance and no fault, are recorded, and the production stopping time of the industrial equipment is recorded. In addition, for an unpredicted fault, operational data is recorded for the five minutes prior to the fault, as well as the time to repair the fault resulting in the shutdown of the industrial equipment.
And step S5, correcting the early warning threshold value of the early warning model according to the early warning information sent by the early warning model, the actual fault of the industrial equipment and the actual production stop time.
As shown in fig. 4, step S5 specifically includes:
step S51, calculating the sum of the production stop time of the prediction error class SumC1 and the sum of the production stop time of the prediction inaccuracy class SumC4 according to the actual production stop time.
Obtaining the sum of the downtime of the predicted error class SumC1 according to the time spent on the downtime maintenance caused by the fact that the early warning model obtained in the step S4 does not predict but actually breaks down within a period of time; meanwhile, the sum of the production stopping time SumC4 of the inaccurate prediction class is obtained according to the time spent in the production stopping maintenance caused by the fact that the early warning model obtained in the step S4 predicts the fault but does not actually generate the fault within a period of time.
In step S52, a weight adjustment constant a is calculated from the sum SumC1 of the downtime of the prediction error class and the sum SumC4 of the downtime of the prediction inaccuracy class.
The sum of the production stopping time of the prediction error class SumC1 and the sum of the production stopping time of the prediction inaccuracy class SumC4 are divided to obtain the ratio a of the two, namely SumC1 and SumC4, and the constant a is a weight adjustment constant. Then SumC1 ═ a × SumC 4.
In step S53, a weight coefficient W1 of the prediction error class and a weight coefficient W4 of the prediction inaccuracy class are obtained according to the weight adjustment constant a.
Weight coefficients are set for centroid values X1 and X4 of the prediction error class C1 and the prediction inaccuracy class C4, and W1 and W4 represent the weight coefficient of the prediction error class and the weight coefficient of the prediction inaccuracy class, respectively. Since it is not desirable that the downtime of the industrial equipment is too long, the longer the downtime, the smaller the weight should be, and then W4 ═ a × W1, and W1+ W4 ═ 1.
From this, W1 is 1/(a +1) and W4 is a/(a + 1).
Since a is a known number calculated by the downtime, W1 and W4 can be calculated.
And step S54, performing cluster analysis according to the early warning information sent by the early warning model and the actual faults of the industrial equipment to obtain a mass center value X1 of a predicted error class, a mass center value X2 of an accurately predicted fault class, a mass center value X3 of an accurately predicted normal class and a mass center value X4 of an accurately predicted inaccurate class.
And step S55, correcting the early warning threshold XT of the early warning model according to the cluster analysis result, the weight coefficient W1 of the prediction error class and the weight coefficient W4 of the prediction inaccuracy class.
The adjustment formula of the early warning threshold XT is as follows:
XT＝(X2+X3)/4+(X4×W4+X1×W1)/2
wherein XT represents an early warning threshold;
x1, X2, X3, and X4 respectively represent a centroid value of a prediction error class, a centroid value of an accurate prediction fault class, a centroid value of an accurate prediction normal class, and a centroid value of a prediction inaccuracy class.
The adjustment of the early warning threshold value improves the accuracy of prediction, reduces the production stop time and improves the production efficiency of industrial equipment.
According to the fault early warning method for the industrial internet, when the early warning threshold value of the early warning model is adjusted, cluster analysis is performed on the prediction result of the early warning model and the actual operation condition of the industrial equipment to obtain the centroid value X1 of the prediction error class, the centroid value X2 of the accurate prediction fault class, the centroid value X3 of the accurate prediction normal class and the centroid value X4 of the prediction inaccuracy class, and meanwhile, the weights of the production stop time caused by the actual fault of the industrial equipment and the production stop time caused by the inaccurate prediction of the early warning model are comprehensively considered, so that the accuracy of early warning is improved, the production stop time caused by the inaccurate prediction is reduced, the production efficiency is improved, unnecessary loss caused by the error of prediction is reduced, the fault early warning and the production efficiency are effectively balanced, and the industrial equipment can safely and efficiently operate; meanwhile, a foundation is laid for realizing real-time updating and standardization flow of decisions.
The embodiment also provides a fault early warning device facing the industrial Internet. As shown in fig. 5, the apparatus includes:
the training model establishing unit 1 is used for training a basic model according to historical production information of the industrial equipment to obtain a training model.
The early warning threshold establishing unit 2 is used for testing the training model to obtain an early warning model and determining an early warning threshold in the early warning model;
the early warning unit 3 is used for early warning actual operation data of the industrial equipment by utilizing an early warning model with an early warning threshold value;
the recording unit 4 is used for recording the early warning sent by the early warning model, the actual fault of the industrial equipment and the actual production stop time;
the early warning threshold correction unit 5 is used for correcting the early warning threshold of the early warning model according to early warning information sent by the early warning model, actual faults and actual production stoppage of the industrial equipment;
the adjustment formula of the early warning threshold value is as follows:
XT＝(X2+X3)/4+(X4×W4+X1×W1)/2
wherein XT represents an early warning threshold;
x1, X2, X3 and X4 respectively represent a centroid value of a prediction error class, a centroid value of an accurate prediction fault class, a centroid value of an accurate prediction normal class and a centroid value of a prediction inaccuracy class;
w1 and W4 respectively represent the weight coefficient of the prediction error class and the weight coefficient of the prediction inaccuracy class; W4/W1 ═ a, and W1+ W2 ═ 1;
a is a weight adjustment constant, and a is SumC1/SumC 4;
wherein SumC1 represents the sum of the downtime of the misprediction class;
SumC4 represents the sum of the downtime of the predicted inaccurate class.
As an optional implementation manner of this embodiment, as shown in fig. 6, the training model establishing unit 1 includes:
and the production information acquisition module 11 is used for extracting historical production information of the industrial equipment.
And the extraction module 12 is used for extracting input data and output data from the historical production information.
A configuration module 13 for configuring the input data and the output data to the base model.
And the training module 14 is used for training the basic model according to the input data and the output data to obtain a training model.
The specific working mode of the training model establishing unit 1 is detailed in step S1 in the fault early warning method for the industrial internet, and is not described herein again.
As another optional implementation manner of this embodiment, as shown in fig. 7, the warning threshold correction unit 5 includes:
the production stop time calculation module 51 is used for calculating the sum of the production stop times SumC1 of the prediction error class and the sum of the production stop times SumC4 of the prediction inaccuracy class according to the actual production stop time;
a weight adjustment constant calculation module 52, configured to calculate a weight adjustment constant a according to the sum SumC1 of the downtime of the prediction error class and the sum SumC4 of the downtime of the prediction inaccuracy class;
and the weight coefficient calculation module 53 is configured to obtain a weight coefficient W1 of the prediction error class and a weight coefficient W4 of the prediction inaccuracy class according to the weight adjustment constant a.
And the cluster calculation module 54 is used for performing cluster calculation according to the early warning information sent by the early warning model and the actual faults of the industrial equipment to obtain a centroid value X1 of a predicted error class, a centroid value X2 of an accurately predicted fault class, a centroid value X3 of an accurately predicted normal class and a centroid value X4 of an accurately predicted inaccurate class.
In the cluster calculating module 54, if the centroid value X4 of the inaccurate prediction class is smaller than the centroid value X1 of the wrong prediction class, or the centroid value X3 of the normal prediction class is smaller than the centroid value X1 of the wrong prediction class, the early warning model is considered to be wrong, and the early warning model is reconstructed.
And the early warning threshold correction module 55 is used for correcting the early warning threshold of the early warning model according to the clustering analysis result, the weight coefficient W1 of the prediction error class and the weight coefficient W4 of the prediction inaccuracy class.
The detailed working steps of the warning threshold correction unit 5 are detailed in step S5 in the fault warning method for the industrial internet, and are not described herein again.
The actual operation flow of the fault warning device provided by the embodiment is briefly described below. As shown in fig. 8, the fault pre-warning device 81 obtains pre-warning model indexes, such as pre-warning threshold values, from the bottom database 83; and then accessing real-time operation data of the industrial equipment 82 and monitoring the real-time operation data of the industrial equipment 82.
When the fault early warning device 81 sends out early warning, a maintenance work order is applied, and meanwhile, an early warning result generation index is sent to the bottom layer database 83 for storage.
After the maintenance work order is generated, the maintenance work order is checked manually, namely, a maintenance manager judges whether maintenance is needed or not according to work experience; if the maintenance is approved, generating a maintenance work order; if the maintenance is not approved, it is considered as a false alarm, and the false alarm and the operation data five minutes before the alarm is given are stored in the bottom-layer database 83.
When a maintenance work order is generated by maintenance management personnel, the maintenance is stopped, and two maintenance structures are provided, wherein one is maintenance completion, namely, a fault is eliminated; another type of ping is completed with no failure. Both cases are stored in the underlying database 83 while downtime is recorded.
The fault early warning device 81 calls the warning data and the manual record of the bottom-layer database, analyzes whether the number of the false early warnings exceeds a preset value, and retrains the early warning model if the number of the false early warnings exceeds the preset value.
The fault early warning device 81 retrieves early warning information stored in a bottom database and actual faults of the industrial equipment, performs cluster analysis, determines an early warning threshold value to balance early warning and capacity, and transmits the early warning threshold value to the fault early warning device 81 to correct the early warning threshold value.
In the fault early warning device for the industrial internet, the early warning threshold correction unit comprehensively considers the weight of the production stopping time caused by the actual fault of the industrial equipment and the production stopping time caused by inaccurate prediction of the early warning model, so that the accuracy of early warning is improved, the production stopping time caused by inaccurate prediction is reduced, the production efficiency is improved, meanwhile, unnecessary loss caused by wrong prediction is reduced, the fault early warning and the production efficiency are effectively balanced, and the industrial equipment is operated safely and efficiently; meanwhile, a foundation is laid for realizing real-time updating and standardization flow of decisions.
It will be understood that the above embodiments are merely exemplary embodiments taken to illustrate the principles of the present invention, which is not limited thereto. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.
Claims (10)
1. A fault early warning method for industrial Internet is characterized by comprising the following steps:
correcting the early warning threshold value of the early warning model according to early warning information sent by the early warning model, actual faults of the industrial equipment and actual production stoppage;
the adjustment formula of the early warning threshold XT is as follows:
XT＝(X2+X3)/4+(X4×W4+X1×W1)/2
wherein XT represents an early warning threshold;
x1, X2, X3 and X4 respectively represent a centroid value of a prediction error class, a centroid value of an accurate prediction fault class, a centroid value of an accurate prediction normal class and a centroid value of a prediction inaccuracy class;
w1 and W4 respectively represent the weight coefficient of the prediction error class and the weight coefficient of the prediction inaccuracy class; W4/W1 ═ a, and W1+ W4 ═ 1;
a is a weight adjustment constant, and a is SumC1/SumC 4;
wherein SumC1 represents the sum of the downtime of the misprediction class;
SumC4 represents the sum of the downtime of the predicted inaccurate class.
2. The method of claim 1, comprising:
training a basic model according to historical production information of industrial equipment to obtain a training model;
testing the training model to obtain an early warning model, and determining an early warning threshold value in the early warning model;
early warning actual operation data of the industrial equipment by using the early warning model with the early warning threshold value;
recording the early warning sent by the early warning model, the actual fault and the actual production stop time of the industrial equipment;
and correcting the early warning threshold value of the early warning model according to the early warning information sent by the early warning model, the actual fault of the industrial equipment and the actual production stopping time.
3. The method of claim 2, wherein the step of training the model based on historical production information of the industrial equipment to obtain a training model comprises:
acquiring historical production information of industrial equipment;
extracting input data and output data from the historical production information;
configuring the input data and the output data to an original model;
and training the original model according to the input data and the output data to obtain the training model.
4. The method of claim 3, wherein the input data comprises: production capacity per unit time, temperature of industrial equipment operation and power consumption;
the output data includes: the data is used for representing the data before the industrial equipment is in failure and representing the data that the industrial equipment is in normal operation.
5. The method of claim 2, wherein the step of correcting the pre-warning threshold of the pre-warning model according to the pre-warning information sent by the pre-warning model, the actual fault of the industrial equipment and the actual production shutdown comprises:
calculating the sum of the production stopping time of the prediction error class SumC1 and the sum of the production stopping time of the prediction inaccuracy class SumC4 according to the actual production stopping time;
calculating a weight adjustment constant a according to the sum of the production stop time SumC1 of the prediction error class and the sum of the production stop time SumC4 of the prediction inaccuracy class;
obtaining a weight coefficient W1 of a prediction error class and a weight coefficient W4 of a prediction inaccuracy class according to the weight adjusting constant a;
performing cluster analysis according to the early warning information sent by the early warning model and the actual fault of the industrial equipment to obtain a mass center value X1 of a prediction error class, a mass center value X2 of an accurate prediction fault class, a mass center value X3 of an accurate prediction normal class and a mass center value X4 of a prediction inaccurate class;
and correcting the early warning threshold value of the early warning model according to the cluster analysis result, the weight coefficient W1 of the prediction error class and the weight coefficient W4 of the prediction inaccuracy class.
6. The method of claim 5, wherein in the step of performing cluster calculation according to the early warning information sent by the early warning model and the actual fault of the industrial equipment,
and if the mass center value X4 of the inaccurate prediction class is smaller than the mass center value X1 of the wrong prediction class, or the mass center value X3 of the normal prediction class is smaller than the mass center value X1 of the wrong prediction class, the early warning model is considered to be wrong, and the early warning model is reconstructed.
7. The utility model provides a fault early warning device towards industry internet which characterized in that includes:
the training model establishing unit is used for training a basic model according to historical production information of the industrial equipment to obtain a training model;
the early warning threshold establishing unit is used for testing the training model to obtain an early warning model and determining an early warning threshold in the early warning model;
the early warning unit is used for early warning actual operation data of the industrial equipment by using the early warning model with the early warning threshold value;
the recording unit is used for recording the early warning sent by the early warning model, the actual fault and the actual production stop time of the industrial equipment;
the early warning threshold correction unit is used for correcting the early warning threshold of the early warning model according to early warning information sent by the early warning model, actual faults and actual production stoppage of the industrial equipment;
the adjustment formula of the early warning threshold value is as follows:
XT＝(X2+X3)/4+(X4×W4+X1×W1)/2
wherein XT represents an early warning threshold;
x1, X2, X3 and X4 respectively represent a centroid value of a prediction error class, a centroid value of an accurate prediction fault class, a centroid value of an accurate prediction normal class and a centroid value of a prediction inaccuracy class;
w1 and W4 respectively represent the weight coefficient of the prediction error class and the weight coefficient of the prediction inaccuracy class; W4/W1 ═ a, and W1+ W4 ═ 1;
a is a weight adjustment constant, and a is SumC1/SumC 4;
wherein SumC1 represents the sum of the downtime of the misprediction class;
SumC4 represents the sum of the downtime of the predicted inaccurate class.
8. The apparatus of claim 7, wherein the training model building unit comprises:
the production information acquisition module is used for extracting historical production information of the industrial equipment;
the extraction module is used for extracting input data and output data from the historical production information;
a configuration module for configuring the input data and output data to the base model;
and the training module is used for training the basic model according to the input data and the output data to obtain the training model.
9. The apparatus of claim 7, wherein the warning threshold correction unit comprises:
the production stop time calculation module is used for calculating the sum of the production stop times SumC1 of the prediction error class and the sum of the production stop times SumC4 of the prediction inaccuracy class according to the actual production stop time;
the weight adjusting constant calculating module is used for calculating a weight adjusting constant a according to the sum of the production stopping time of the prediction error class SumC1 and the sum of the production stopping time of the prediction inaccuracy class SumC 4;
the weight coefficient calculation module is used for obtaining a weight coefficient W1 of the prediction error class and a weight coefficient W4 of the prediction inaccuracy class according to the weight adjusting constant a;
the cluster calculation module is used for carrying out cluster calculation according to the early warning information sent by the early warning model and the actual faults of the industrial equipment to obtain a mass center value X1 of a predicted error class, a mass center value X2 of an accurate predicted fault class, a mass center value X3 of an accurate predicted normal class and a mass center value X4 of an inaccurate predicted class;
and the threshold correction module is used for correcting the early warning threshold of the early warning model according to the cluster analysis result, the weight coefficient W1 of the prediction error class and the weight coefficient W4 of the prediction inaccuracy class.
10. The apparatus of claim 9, wherein in the cluster computation module,
and if the mass center value X4 of the inaccurate prediction class is smaller than the mass center value X1 of the wrong prediction class, or the mass center value X3 of the normal prediction class is smaller than the mass center value X1 of the wrong prediction class, the early warning model is considered to be wrong, and the early warning model is reconstructed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910372649.9A CN110334728B (en) | 2019-05-06 | 2019-05-06 | Fault early warning method and device for industrial internet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910372649.9A CN110334728B (en) | 2019-05-06 | 2019-05-06 | Fault early warning method and device for industrial internet |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110334728A CN110334728A (en) | 2019-10-15 |
| CN110334728B true CN110334728B (en) | 2022-04-01 |
Family
ID=68139613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910372649.9A Active CN110334728B (en) | 2019-05-06 | 2019-05-06 | Fault early warning method and device for industrial internet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110334728B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| SG11201708172RA (en) * | 2015-04-07 | 2017-11-29 | Tlv Co Ltd | Maintenance support system and maintenance support method |
| CN110861987B (en) * | 2019-10-23 | 2021-06-08 | 日立楼宇技术（广州）有限公司 | Elevator fault judgment logic verification method, system and storage medium |
| CN111044808A (en) * | 2019-11-15 | 2020-04-21 | 国网江苏省电力有限公司 | Power utilization information acquisition system operation and maintenance quality reliability assessment system and method |
| CN111241688B (en) * | 2020-01-15 | 2023-08-25 | 北京百度网讯科技有限公司 | Method and device for monitoring composite production process |
| CN113205666B (en) * | 2021-05-06 | 2022-06-17 | 广东鹰视能效科技有限公司 | Early warning method |
| CN113341774B (en) * | 2021-05-31 | 2021-12-28 | 浙江锐博科技工程有限公司 | Large-scale public building energy consumption monitoring system |
| CN116339267B (en) * | 2023-05-25 | 2023-08-08 | 深圳市星火数控技术有限公司 | Automatic production line control system based on Internet of things |
| CN117369392B (en) * | 2023-11-17 | 2024-04-16 | 岳阳长炼机电工程技术有限公司 | Equipment fault intelligent early warning method based on multiparameter logic relation |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491375A (en) * | 2017-08-18 | 2017-12-19 | 国网山东省电力公司信息通信公司 | Equipment detection and fault early warning system and method under a kind of cloud computing environment |
| CN108241901A (en) * | 2016-12-27 | 2018-07-03 | 全球能源互联网研究院 | A kind of transformer Forewarn evaluation method and device based on prediction data |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104020751B (en) * | 2014-06-23 | 2016-08-24 | 河海大学常州校区 | Campus Security monitoring method based on Internet of Things |
| CN107704495B (en) * | 2017-08-25 | 2018-08-10 | 平安科技（深圳）有限公司 | Training method, device and the computer readable storage medium of subject classification device |
| CN107947156B (en) * | 2017-11-24 | 2021-02-05 | 国网辽宁省电力有限公司 | Power grid fault critical clearing time discrimination method based on improved Softmax regression |
| CN109636787B (en) * | 2018-12-11 | 2022-09-16 | 哈尔滨工业大学 | High-precision real-time battery spot welding quality detection method based on deep learning |
| CN109670551A (en) * | 2018-12-20 | 2019-04-23 | 树根互联技术有限公司 | The failure prediction method and device of engineering mechanical device |
-
2019
- 2019-05-06 CN CN201910372649.9A patent/CN110334728B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108241901A (en) * | 2016-12-27 | 2018-07-03 | 全球能源互联网研究院 | A kind of transformer Forewarn evaluation method and device based on prediction data |
| CN107491375A (en) * | 2017-08-18 | 2017-12-19 | 国网山东省电力公司信息通信公司 | Equipment detection and fault early warning system and method under a kind of cloud computing environment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110334728A (en) | 2019-10-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110334728B (en) | Fault early warning method and device for industrial internet | |
| CN104809051B (en) | Method and apparatus for predicting exception and failure in computer application | |
| CN111209131A (en) | Method and system for determining fault of heterogeneous system based on machine learning | |
| CN103754718B (en) | Lift running safety monitoring system and method thereof | |
| CN109884473B (en) | Electric power overhaul system and method | |
| CN106155035B (en) | Method for diagnosing faults and fault diagnosis system based on repair class data | |
| CN107271829A (en) | A kind of controller switching equipment running state analysis method and device | |
| CN110865924B (en) | Health degree diagnosis method and health diagnosis framework for internal server of power information system | |
| CN115225463B (en) | Hardware fault monitoring method and system based on IT operation and maintenance | |
| CN114862275B (en) | Storage logistics system reliability assessment method and system based on digital twin model | |
| CN113448298A (en) | Data acquisition system for automatic production equipment | |
| CN107247198A (en) | A kind of distribution equipment malfunction Forecasting Methodology and device | |
| CN107103425B (en) | Intelligent energy evaluation system for power generation equipment running state computer | |
| RU2687848C1 (en) | Method and system of vibration monitoring of industrial safety of dynamic equipment of hazardous production facilities | |
| CN116910680B (en) | Remote fault detection method and system for fitness equipment | |
| CN115114124A (en) | Host risk assessment method and device | |
| CN112381242A (en) | Nuclear power station equipment maintenance project data processing method and system | |
| CN112346893A (en) | Fault prediction method, device, terminal and storage medium | |
| CN111210092A (en) | Stacking machine predictive maintenance method and system based on deep learning | |
| TWI794907B (en) | Prognostic and health management system for system management and method thereof | |
| CN109754490A (en) | Health control method, device, medium and the equipment of brake system of electric locomotive | |
| CN115600695A (en) | Fault diagnosis method of metering equipment | |
| CN112836380A (en) | Fault estimation and maintenance method for intelligent manufacturing production line | |
| CN113240358B (en) | Automatic recording method for creep data of ultra-large span cable-stayed bridge construction | |
| CN117272844B (en) | Method and system for predicting service life of distribution board |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |